PROMPT GUIDE

How to Use ComfyUI for Beginners (Nodes and Your First Workflow)

ComfyUI is one of the interfaces for running Stable Diffusion-style models. Instead of a screen full of buttons, you connect processing boxes (nodes) with wires and decide yourself what happens in which order. It looks hard at first, but the basic setup is only six nodes. This guide explains those six, the steps to your first image, and how to fix the errors beginners hit most.

Last updated: 2026-10-22 approx. 4 min read

日本語版

The short answer

ComfyUI is a Stable Diffusion interface where you build the image-making process by connecting boxes called nodes. The fastest way in is to run the default workflow as it is (load model, prompt, sampler, decode, save) and change only the prompt and image size at first.

What is ComfyUI? (An interface built from nodes)

It is a Stable Diffusion interface that splits image generation into boxes (nodes) and sets the order by wiring them together.

Interfaces like AUTOMATIC1111 or Forge show a prompt box, size fields and a Generate button from the start. In ComfyUI, the steps behind those fields, such as loading the model, turning text into numbers, building a picture from noise and decoding it into an image, are each a separate box, and the wiring decides the process. Its strengths are that you can save a workflow and reuse the exact same setup, and that you can keep adding finer steps.

Point	ComfyUI	Button-style UI (Forge etc.)
Look	Boxes connected by wires	Tabs and input fields
First hurdle	Higher (learn what the wires mean)	Lower (fill in fields and click)
Saving your setup	Save and load whole workflows	Save some settings
Fine control	Add nodes and build freely	Install extensions

Anime girl in a school uniform smiling under cherry blossoms, an example of a first image — 最初の1枚の例

Full-body anime girl in a tall portrait size, an example of choosing image size — 最初の1枚の例

Illustration usage rules

Which nodes do you need for your first image? (The basic six)

Load Checkpoint, two text prompts, an empty latent, the sampler, VAE Decode and Save Image are enough for one picture.

Node (common name)	What it does	What you change
Load Checkpoint	Loads the model (checkpoint)	Pick the model
CLIP Text Encode (x2)	Turns the prompt and negative prompt into numbers	Write your text or tags
Empty Latent Image	Prepares a blank base for the image	Width, height, batch size
KSampler	Builds the picture from noise step by step	seed, steps, cfg, sampler
VAE Decode	Turns the working data back into a normal image	Usually nothing
Save Image	Saves the image	Filename prefix

Rinka's tips

Node names and looks can change a little between versions. If you are unsure, check the official docs or use the node search
Wires normally connect sockets of the same color. Different colors will not connect
Use two CLIP Text Encode nodes: the top one for the prompt, the bottom one for the negative, wired into the positive and negative inputs of KSampler

How do you make your first image in ComfyUI?

Load the default workflow, pick a model, write a prompt and run it.

Steps to your first ComfyUI image

Put the model file you want to use into the checkpoints folder inside ComfyUI's models folder
Start ComfyUI and open the address it shows in your browser
Use the default workflow that appears at start (if you deleted it, load the default workflow again from the menu)
In Load Checkpoint, choose the model you added
Write the prompt in the top CLIP Text Encode and the negative prompt in the bottom one
Set the width and height in Empty Latent Image to a size that suits the model (for SDXL-type models, for example 832x1216)
Press the run button (named Queue or similar) and check the image that appears

1girl, solo, smile, long hair, school uniform, cherry blossoms, outdoors, upper body, masterpiece, best quality

worst quality, low quality, bad anatomy, bad hands, extra fingers, blurry, text, watermark

For a list of SDXL sizes to use in Empty Latent Image, see SDXL Resolutions (Recommended Sizes) and How to Pick a Ratio →

Where do models and LoRAs go?

Inside ComfyUI's models folder there is a folder for each type of file.

File	Example folder	Note
Checkpoint (main model)	models/checkpoints	Appears in Load Checkpoint
LoRA	models/loras	Add a LoRA loader node to use it
VAE	models/vae	Use a different VAE when colors look washed out
Upscaler	models/upscale_models	Used by upscaling nodes
ControlNet	models/controlnet	For fixing pose or line art

After adding files, refresh the page or restart and they will appear in the lists. Folder names can be added or changed between versions, so also check the included README and the official docs.

For LoRA strength and how to use LoRAs, see How to Use LoRA (Syntax, Strength and Trigger Words) →

How do you save and share a workflow?

Workflows can be saved as JSON, and PNG images made in ComfyUI often contain the workflow too.

Dragging a PNG saved by ComfyUI onto the canvas often opens the exact workflow that made it. This is handy for trying workflows other people share, but some need custom nodes you do not have. When you give images to others, remember the prompt and workflow may be inside; re-save the image to strip the metadata if needed.

What if you get red nodes or errors in ComfyUI? (Common mistakes and fixes)

Most errors are a missing file, a missing custom node or not enough memory.

Problem	Cause	Fix
A node turns red	A custom node the workflow needs is not installed	Install the missing nodes (for example with a custom node manager) and restart
Model cannot be selected or found	Wrong folder, or the list was not refreshed	Put it in the correct models folder and refresh
Stops midway or shows out of memory	Image too large or too many at once	Reduce the size, set batch to 1, close other apps
Black or grey image	Wrong VAE or extreme settings	Use a VAE that matches the model, return cfg and steps to normal values
Wired up but nothing runs	One wire is missing	Check that positive, negative, model and latent are all connected

Rinka's tips

Copy and keep the error text; it makes the cause much easier to find
Before using someone else's workflow, confirm you can make one image with the default workflow
Custom nodes are useful, but too many slow down startup or conflict. Keep only what you need

If colors look dull or grey, read about the VAE in What Is a VAE? Why Colors Look Dull, and How to Choose and Set a VAE →

To control pose or line art, see how to use ControlNet in How to Use ControlNet (OpenPose, Canny, Depth and Lineart Compared) →

If you prefer a button-style interface, see the Forge guide in How to Use Stable Diffusion Forge (Reading the UI and Making Your First Image) →

FAQ

Frequently Asked Questions

Can beginners use ComfyUI?

Yes. Start by running the default workflow as it is and change only the prompt and size; that keeps things simple. Add nodes once you are comfortable.

ComfyUI or Forge (WebUI): which is better?

If you want pictures right away, a button-style interface like Forge is easier. ComfyUI suits you if you want to build the process in detail or reuse the same process many times. Many people use both for different jobs.

My nodes are red and nothing runs.

Usually a custom node the workflow needs is not installed. Install what is missing and restart; if that does not fix it, also check where the model is and what it is called.

Does the prompt stay inside images made with ComfyUI?

Saved PNGs often contain the workflow and prompt. Check before sharing, and re-save the image to remove the metadata if you do not want to leave it.